Fault diagnosis method based on fault data deep mining and learning

Abstract

The invention belongs to the technical field of generator set equipment fault diagnosis, in particularly to a fault diagnosis method based on fault data deep mining and learning, which comprises the steps of collecting and preprocessing the historical data of a generator set, then learning and training through a deep long-short time memory network algorithm, obtaining a fault data screening model, and then traversing a large number of historical databases, and screening and forming a fault data sample set; estimating the number of the fault types of the fault data sample set by adopting a Medoids surrounding classification method, and adopting K-Means clustering algorithm for clustering analysis to form a multi-class typical fault sample set; training and learning the multi-class typical fault sample set by adopting an LSTM neural network algorithm, and establishing a fault diagnosis model; and monitoring the real-time operation data of the system based on the fault diagnosis model, judging the operation state of the system, recording a newly generated fault sample, and updating the fault diagnosis model by using the updated multi-class typical fault sample set.

Description

technical field [0001] The invention belongs to the technical field of fault diagnosis of generating set equipment, and in particular relates to a fault diagnosis method based on deep mining and learning of fault data. Background technique [0002] Equipment fault diagnosis technology is very important to the daily safety of industrial equipment. It is not only related to daily production arrangements and equipment maintenance, but also can detect hidden faults of equipment and eliminate them in time to avoid sudden failures of important equipment from affecting the overall safety and stability of the production process. . The premise of fault diagnosis is the monitoring and analysis of equipment state quantities. Taking power plants as an example, monitoring systems in power plants are widely used, and are equipped with historical databases that can store massive operating data of power plants. Power plant operation data is rich in content, and several months of data can o...

AI Technical Summary

Problems solved by technology

Power plant operation data is rich in content, and several months of data can often cover all working conditions of equipment operation. However, the data storage of power plants is huge and difficult to analyze. The data mining method combined with deep learning technology that has emerged in recent years can effectively solve this problem. a question

[0003] The essence of fault diagnosis and identification methods is the classification and regression problems in data mining and deep learning. In the past, fault diagnosis methods commonly used methods such as Fourier transform, wavelet transform, statistical analysis, and spectrum analysis. These processing methods often rely on signal processing technology and Diagnosis experience, manual extraction of fault features is cumbersome and complicated, which increases the complexity of diagnosis

Images